CMS collection keep working during holiday
Peter B. Kessler
Peter.Kessler at Sun.COM
Thu Oct 9 10:18:29 PDT 2008
Would it be diagnostic to get a "jmap -histo" right after a CMS cycle
completes (e.g., by watching the GC log), followed immediately by a
"jmap -histo:live" to force a full compacting GC? Differencing those
might tell you what CMS isn't seeing. There will be some "fuzz" in the
result because the application continues running and allocating
short-lived objects, e.g., during the concurrent mark and concurrent
sweep, and which the full collection will collect, but maybe someone who
knows the application could see past the differences in short-lived stuff.
Tony Printezis wrote:
> Y Srinivas Ramakrishna wrote:
>> If you use jmap -histo:live <pid> from the command-line you will
>> get a full gc before the histogram is computed. This full gc will
>> be done despite having +DisableExplicitGC.
> In fact, given that we want to find out what's growing in the heap
> during the CMS cycles (which seems to be GCed during Full GCs), it's
> better to use jmap -histo (i.e., no :live) to avoid the Full GC. But I'm
> forgetting: which VM Ken using? IIRC only jmap in 6 has the :live / no
> :live functionality (before that, it always did a Full GC, which is not
> what you want some of the time).
More information about the hotspot-gc-dev